Roux (French pronunciation: [ʁu]) is a Rubik's cube speedsolving method invented by Gilles Roux. Unlike the CFOP and Petrus methods, the inventor of this method has used it to achieve an official sub-15 average. Similarities can be drawn to the Petrus method's block building and the Waterman method's layer-on-the-left and edges-last aspects.
1. Build a 1x2x3 block anywhere on the cube.
2. Build a second 1x2x3 block opposite of the first 1x2x3 block, without disrupting the first 1x2x3 block. After this step, there should be two 1x2x3 blocks: one on the lower left side, and one lower right side, leaving the U slice and M slice free to move.
4a. Orient the 6 remaining edges using only M and U moves (UF, UB, UL, UR, DF, DB need to be oriented correctly).
4b. Solve the UL and UR edges, preserving edge orientation. After this step, both the left and right side layers should be complete.
4c. Solve the centers and edges in the M slice, see Last Six Edges.
- See also: L5E
- Like the Petrus method, the Roux method uses fewer moves than the popular Fridrich method.
- It is also more intuitive and requires fewer algorithms.
- After the first block is built the rest of the cube can be solved mostly with R, r, M and U moves thus eliminating rotations.
- CMLL is one of the best algorithm sets as there are only 42 cases and most algorithms are fast OLLCPs from CFOP
- The blockbuilding and intuitive nature of the method allows for rapid improvements in lookahead and inspection
- The LSE step of Roux is very easy to master, as it has easy lookahead and allows for fast, 2-gen MU TPS.
- Block building can be difficult for a beginner to get used to. The reliance on r and M moves may also be difficult for some people, so much so that cubers who have trouble with M turns should probably not use this as their main method (or better, practice the M moves). Slice turns can also be slower than outer turns.
- Since the M-slice is used often, especially in the final stages, there is a larger chance of a DNF rather than a +2 if the solver misses the second flick in an M2, or if the solver misses the last M move. It is a DNF because M uses both the R and L face in one.
- The M-slice becomes increasingly difficult with higher order puzzles. With 7x7x7 and 6x6x6, many argue that Roux is essentially unusable; however, with practice, one may be able to do well with it on big cubes.
- The M-slice is very difficult with OH, and OH Roux solvers almost always need to utilize table abuse and therefore can pretty much never solve one-handed away from a table or other surface. However, if one practices doing one-handed M-slices, it can be done very well but only with a table (which shouldn't be a concern in competitions).
Free/non-linear blocks: The first block and second block do not need to be built in that order. You can build part of one and finish the other later. This is very useful when there are a lot of free, pre-built blocks and pairs.
Non-matching centres: The first two blocks can be built around incorrect centres. This allows for more efficiency and allows rouxers to take advantage of pre-built blocks. The centres are corrected directly before CMLL with either u M' u' or u' M' u.
Some Roux users have learned multiple algorithms for each CLL case, each affecting edges in a different way. This allows for manipulation of edge orientation, leading to an easier LSE. This is called CMLLEO.
Possibly another improvement is to expand to solving any of the four second blocks. This means that the D-layer colors of the two blocks don't have to match. This is called non-matching blocks. One negative, if someone has already learned a recognition method for CLL, is that to easily use the full range of second block options, a switch to NMCLL recognition is necessary. Another is that LSE EO is a bit more difficult to recognize in some solves.
Pinkie Pie: a variant proposed by Alex Lau in 2016, it involves orienting the UL/UR edges on the D layer while using an OLLCP algorithm to orient the remaining edges and solving the remaining corners. The user then will get a 4a skip and a very easy 4b step. While this may seem good, many faster Rouxers are of the opinion that it is simply better to influence the EO step and not go to all the hassle of placing UL/UR on D and having to recognise an OLLCP. There is also the reason that many Roux solvers use the method to have less algorithms and do not want to learn the huge OLLCP algset.
EOLR: a variant proposed by Gilles Roux and further documented by Alex Lau, Kian Mansour, and Iuri Grangeiro. It combines steps 4A and 4B of the method, orienting edges and placing ULUR on the D layer simultaneously. The cases are all intuitive, as they involve performing the EO case from different angles to set up a "good arrow" where ULUR can then be placed on bottom.
UFUB: Instead of solving ULUR edges in 4B, UFUB are solved. This can lead to more efficient solutions, but lookahead becomes much more difficult. It is most useful for skipping the "dots" 4C case.
Misoriented centres: Standard roux involves orienting all 6 edges relative to the centre colour that is on the bottom of the blocks. Instead, we can orient them relative to the front colour. As with UFUB, efficiency is improved, but lookahead is hindered. This is most useful for known EOLR cases.
- Gilles Roux 
- Thom Barlow (Kirjava) 
- Austin Moore (BigGreen) 
- Jules Manalang (Waffo) 
- Alexander Lau (5BLD) 
- Kaijun Lin (林恺俊) 
- Mitsuki Gunji
- Artur Kristof (Arcio) 
- Kian Mansour 
- Kavin Tangtartharakul (GuRoux) 
- Vincent Wong (Kangaroux) 
- Roux help and discussion thread
- One-question answer thread
- Reading Roux corners tables
- Beginner's Roux video tutorial
- Roux Example solves
- Roux step 4b-4c transition help